[PATCH 4.9] bpf/verifier: Fix states_equal() comparison of pointer and UNKNOWN

[Ben Hutchings <ben@decadent.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 1368 bytes --]

An UNKNOWN_VALUE is not supposed to be derived from a pointer, unless
pointer leaks are allowed.  Therefore, states_equal() must not treat
a state with a pointer in a register as "equal" to a state with an
UNKNOWN_VALUE in that register.

This was fixed differently upstream, but the code around here was
largely rewritten in 4.14 by commit f1174f77b50c "bpf/verifier: rework
value tracking".  The bug can be detected by the bpf/verifier sub-test
"pointer/scalar confusion in state equality check (way 1)".

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Cc: Edward Cree <ecree@solarflare.com>
Cc: Jann Horn <jannh@google.com>
Cc: Alexei Starovoitov <ast@kernel.org>
---
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -2722,11 +2722,12 @@ static bool states_equal(struct bpf_veri
 
 		/* If we didn't map access then again we don't care about the
 		 * mismatched range values and it's ok if our old type was
-		 * UNKNOWN and we didn't go to a NOT_INIT'ed reg.
+		 * UNKNOWN and we didn't go to a NOT_INIT'ed or pointer reg.
 		 */
 		if (rold->type == NOT_INIT ||
 		    (!varlen_map_access && rold->type == UNKNOWN_VALUE &&
-		     rcur->type != NOT_INIT))
+		     rcur->type != NOT_INIT &&
+		     !__is_pointer_value(env->allow_ptr_leaks, rcur)))
 			continue;
 
 		/* Don't care about the reg->id in this case. */

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 811 bytes --]